Loss Functions - EXPLAINED!

[@sephchan50]

'check it out when you want to lower your self esteem' damn that sentence caught me off guard

[@BB-rh5bp]

Nice voice, tones, speed, and graphics! Perfect for studying anytime. Thanks!

[@oscarsal433]

Wow impresive more value than dozens of study hours!!! Love it!

[@CodeEmporium]

Im Glad this helps people like you :)

[@MarcelloNesca]

This was an incredibly accessible video! You explained complex subjects simply enough to make me curious about going deeper in loss functions. Very awesome, thank you

[@CodeEmporium]

Perfect! That's the intention :)

[@kayingtang1493]

Finally find a video that explains loss function clearly. Thanks for your effort!!

[@CodeEmporium]

Glad it helps!

[@bhargav7476]

Welsch loss is what I needed for my problem, thanks

[@geeky_explorer9105]

Mannnnn! you nailed it, thanks a lot for this video, great informative lecture, keep making this contents and keep inspiring us

[@mahaalabduljalil6596]

I enjoyed the video so much but honestly subscribed to keep the lights ON in your apartment :)

[@hillosand]

Absolute loss does not 'ignore outliers altogether'. Outliers still have more effect on the absolute loss function than non-outliers do, it's just not as extreme an effect as MSE. The right-hand graphic at about 2:35 is wrong, the fitted line would be shifted much more upwards for low values of x.

[@handokosupeno5425]

Amazing explanation bro

[@ShouseD]

Awesome explanation!

[@GauravSharma-ui4yd]

Hey ajay this is gaurav, I always loved to watch your in depth videos. Please make indepth videos on various optimizers like adam, adagrad, ftrl and all other

[@CodeEmporium]

Thanks! Thinking about that optimizer video.

[@AhmedThahir2002]

How is Squared Loss equivalent to setting alpha=2 in the adaptive loss? Won't alpha=2 make the numerator as 2-2=0?

[@1337Rinz]

bro you are my hero 🥰

[@emrek1]

You said pseudo-huber loss is differentiable at 3:00 but it is not continuous if alpha is not 1. It may not be differentiable even when alpha is 1, am I wrong?

[@satyendrayadav3123]

You are awesome man!!!

[@templetonpeck1084]

This video was the was freaking awesome!

[@Manicscitzo]

Well explained! Good work

[@CodeEmporium]

Thanks!

[@gregorygreif2140]

I love your animations! How did you make them?

[@KaranKinariwala1995]

Hey, firstly great explanation! Thanks! Secondly, I have a minor clarification. Can we then, in conclusion, say that finding the best loss function for a particular task also depends on the second derivative of the loss function? Since the final loss depends on the convex nature of the loss function.

[@vivek2319]

Thanks for this video, nicely explained.

[@emmanuelgoldstein3682]

Hello, not sure if you will reply to this older video but at 7:01 you choose the function with the greatest amount of loss and say that it fits the data the best. Was this an error? If not, can you explain why?

[@skyp4216]

I beleive its because tge outliers affect is the least

[@ethiopiansickness]

Great video! But I really need help understanding something. I thought the purpose of calculating MSE, SSE etc. Was to get a sense of accurate your regression model is. 0:55 I noticed how when you introduced the outliers you said that the model tries to change to also incorporate the outliers. So does that also change the linear regression model too? I'm also confused because I thought linear regression (OLS) already computes the minimum of the sum of squared errors.

[@codingwithnipu]

what tool you used for the visulization graph

[@jonathanduran2921]

One thing I never understood.. why are we calling the L2 error the sum of squares.. there’s a square root that is implemented on the sum of the squares.. doesn’t this offset the impact of outliers on the L2 norm?

[@ankitjaiswal272]

Best Part: Check it out to lower your self esteem 😂😂

[@CodeEmporium]

Thank you for noticing the humor haha

[@johnnysmith6876]

NICE video 😀😀

[@hariharans.j5246]

Great as always man! Make videos about reward modelling and neural ODEs

[@CodeEmporium]

Thanks homie. I'll check more on Neural ODEs

[@menglin7432]

This is gold

[@euclitian]

do you use 3b1b's manim library to make the animations?

[@CodeEmporium]

I only used manim for a gradient descent video I made some time ago. I used a lot of Jonathan Barron's animations here (description for more details)

[@3washoka]

Thank you for the video

[@Slim-bob]

sooooo goood!

[@students4life821]

Your video is great but the intro scares me :p Thanks, mate!

[@EW-mb1ih]

What is epsilon insensitive loss function?

[@ankitganeshpurkar]

good info

[@CodeEmporium]

Thanks for watching

[@nadiaterezon8281]

Sorry I feel confused... how do we even calculate the absolute loss and squared loss in the first place

[@hardikvegad3508]

One question how do we apply/implement adaptive loss? do we have to create a list of random values of alpha; like we do in randomizedsearchcv or grid search. Or is there any other way?.... Btw your explanation was amazing.

[@arkasaha4412]

Nice video! How about doing a mathematical video related to eigenvalues/vectors in ML applications?

[@DiogoSanti]

Great job!!!!

[@alexandermass4552]

As a physicist I call this a potential . think also mathematicians would call it as such. only people at Google and who think artificial intelligence has something to do with commercial industry do not understand that it was originally a matter of science.

[@CodeEmporium]

True. Many industry practitioners are interested in “How can I used AI to help me” over “What is AI”

[@gilbertcabigasalmazan3289]

How do you visualize data points?

[@davidmurphy563]

"Check it out if you want to lower your self-esteem" My self-esteem is outputted by a ReLU activation function, you can't touch it.

[@david-vr1ty]

Great video. Is the code for the graphics somewhere available? I would like to use and adapt it for my purposes

[@fernandogonazales1797]

great vid ty

[@CodeEmporium]

Anytime!

[@ehza]

nice man

[@Usernotknown21]

Maybe start off with what is a loss function

[@GaneshMohane]

Good

[@arsalan2780]

can u recommend books as well from where you read this.. i know couple but it covers all topics of DL

[@reynaldoquispe9335]

good explains (Y)

[@jameswood6445]

10/10

[@piqieme]

i was forced to study about this by myself for my final project , even though im major in networking. ughhh.

[@CodeEmporium]

I know. It sucks, right?

[@piqieme]

@@CodeEmporium it sure is. All my friends are working on IOT, and me? damn. Deep learning. haven't been able to sleep normally for this past few months learning everything from scratch.

[@haimantidas843]

is this video for ghost only?

[@haimantidas843]

😄

[@revimfadli4666]

Is this loss?

[@CodeEmporium]

Yea. Different types of loss functions you can optimize for when creating a model

[@shmarvdogg69420]

why is this video second to siraj ravals? his is prolly a rip off of this one so really you should be on top

[@vinayreddy8683]

*why don't you upload 2-3 videos per week* just saying.

[@sushantkarki2708]

too much information to process at once

[@arturr5]

This didn't explain anything. in cross entropy you used a comparison to a cell tower but didn't translate it to machine learning at all. In a real model what are the 3 bits? what are the 5 bits??????

[@menaxyzhou9479]

lmao 6:03

[@maltml]

Isn't adaptive loss kind of... dumb? Instead of solving a problem you redefine the problem until you get a satisfactory answer. And apart from 2D regression you have no way of knowing if model is doing what you want.

[@johnheikens]

you just copied the video from jon Barron and added your own voice!

[@CodeEmporium]

I used his animations for 2 of the losses and the comparison of 4 losses later in the video. Attributed him when I talk about his paper and the top of the description. Also, 4 of the 6 losses discussed were NOT in his video. He never talked about classification or the absolute loss.

[@johnheikens]

@@CodeEmporium I hope you asked his permission (as this video has potential to replace his video). i see that you have a creative commons license on your video, while his video doesn't. This means that if someone uses your video, because he would think it is creative commons, he also gets a copystrike as soon as Jon Barron copystrikes your video. You cant just copy parts of another video, and tell people they are fine when they use it.(by cc license)

[@thedailyepochs338]

@@johnheikens hey john

[@thedailyepochs338]

@@johnheikens Shut up

[@Youshisu]

half of your video is stolen xD